Aprendizado profundo: redes LSTM

Abstract

Over the last decade, there has been a very rapid advance in neural networks research. Due to the success of new techniques, big results in competitions in several fields, and its increasing use in industry, this field gained its own name, which has become a buzzword over the past years: deep learning. This work seeks to offer a brief introduction on deep learning, exploring its techniques and explaining what constitutes a deep neural network. After this brief introduction, it will be discussed in more detail one of the techniques responsible for the recent success of deep learning, the Long Short Term Memory (LSTM) networks, and how this technique can be applied to time series prediction. From the results obtained training a LSTM network for time series prediction using four datasets, the LSTM network performed better in three of these sets compared to other methods, which shows their effectiveness against other already consolidated techniques for this kind of problem, such as artificial neural networks and ARIMA.